e2e

README

End-to-end harness

Drives spaceship as real processes — a server binary and a client binary talking over a real gRPC tunnel — and exercises it through the front ends an operator actually uses.

The in-tree internal/e2e package deliberately covers one leg at a time: the router is process-global, so a single test process cannot route a destination to the proxy egress on the client side and to direct on the server side. Separate processes are the only way to cover both legs at once, which is what this harness does.

go build -o /tmp/spaceship ./cmd/spaceship && go run ./scripts/e2e /tmp/spaceship /tmp/spaceship-e2e

Arguments are the binary under test and a scratch directory for configs, keys, and per-process logs. Exit status is non-zero if any check fails.

Coverage:

• h2c and TLS tunnels: 4 MiB SOCKS5 round trip verified by SHA-256, 40 concurrent tunnels, HTTP CONNECT, absolute-form HTTP GET, SOCKS5 UDP associate.
• basic_auth accepted and rejected.
• Shutdown under load: 40 established sessions plus a peer frozen with SIGSTOP mid-session, then SIGTERM. Both processes must exit within 2s. This is the regression that shipped in v2.1.7, where the server called grpc GracefulStop first and waited on peers that never answered the GOAWAY.
• Restart cycle: stop the server, rebind the port, and confirm a still-running client recovers on its own.
• -stop-timeout watchdog stays silent on a healthy stop.

Documentation

Overview

Command e2e drives spaceship end to end as real processes: a server binary and a client binary talking over a real gRPC tunnel, exercised through the front ends an operator actually uses.

The in-tree e2e package cannot do this — the router is process-global, so one test process cannot route a destination to the proxy on the client side and to direct on the server side. Separate processes is the only way to cover both legs at once.